Change Detection with LiDAR Data David Streutker Idaho State University Boise Center Aerospace Lab Naval Postgraduate School LiDAR Littoral Studies Workshop Monterey, California May 24, 2007 Introduction • • • • • • Uses of change detection LiDAR accuracy and change detection Two methods of co-registration Co-registration example Change detection of a landslide Change detection of a rangeland fire Uses of Change Detection • Coastal studies – Beach erosion and/or deposition • Hydrology – Water levels (surface and groundwater) – Snow pack – Bathymetry • • • • Aeolian transport Tectonic and landslide movement Volcanology Vegetation monitoring LiDAR Accuracy • Accuracy determines amount of change detection possible • Absolute accuracy – Accuracy with respect to global coordinate system – Generally around 15 cm vertical and 50 cm horizontal • Relative accuracy – Accuracy within dataset (“point-to-point”) – Can be better than 5 cm vertical Co-Registration • Necessary for change detection • “Brute force” method – Use of least-squares to evaluate fit – Iterative to determine best fit – Computationally expensive • Slope-based method – Intelligent • Estimates overall offset – Flexible • Able to use polynomial warping – Computationally efficient Salmon Falls Creek Landslide Salmon Falls Creek Landslide • Data acquired in 2002 and 2005 • 2002 data – 1 m spacing – High relative accuracy (< 25 cm vertical) – NAD 27 datum • 2005 data – 0.5 m spacing – Very high relative accuracy (< 10 cm vertical) – NAD 83 datum Accuracy of 2002 Data • Relative accuracy “poorer” than 2005 data • Primary reason due to small errors in flightline co-registration – Difficulty due to rugged terrain • Problem: Relative accuracy on the order of or lower than the expected change • Solution: Redo flightline co-registration Flightline Overlap Analysis 0 + Vertical Difference Example: No Offset Example: X Offset = 0.5 m Example: X Offset = 1 m Example: X Offset = 2.5 m Example: X Offset = 5 m X Offset = 1 m, Z Offset = 1 m Vertical Offset Versus Slope • • • • Linear relationship implies shift Shift amount in X and Y can be determined by the slope Offset in Z determined from flat regions Surfaces can be corrected by shifting in X, Y, and Z Data Density Before and After Correction Before After Average Offsets Before After • Vertical offset measured from flat areas • Horizontal offset measured from steep areas • Distribution of offsets provides measure of relative accuracy Recent Landslide Activity Components of the slide Landslide Change Detection • Co-registration technique applied to 2002 and 2005 datasets – Areas of known change were masked to avoid bias • Landslide • Ponds and lakes • Quarry – Used a robust, least absolute deviation to avoid bias from outliers • Co-registered images were subtracted from one another to determine change Comparison of Profiles Overall Vertical Shift -0.5 0 Vertical Difference (m) +0.5 Deconvolving Horizontal Movement X Offset = -78 cm Z Offset = 12 cm U.S. Sheep Experiment Station • Near Dubois, Idaho • Vegetation heights of 50 - 150 cm • Major Species – Mountain Sagebrush, Rabbitbrush, Horsebrush – Thickspike wheatgrass, Plains reedgrass, Idaho fescue USDA Sheep Experiment Station • A prescribed burn took place in the fall of 2005 • LiDAR data were acquired in the weeks before the burn, and again soon after the burn • Vegetation heights were determined from both the pre- and post-burn data • Surface texture products were compared to estimate burn severity Vegetation Roughness: 1D Vegetation Roughness: 2D Before 0 After Vegetation Roughness (cm) 20 Vegetation Change • Clear burn signature • Variations in the amount of change indicate burn severity 0 Decrease in Roughness (cm) 15 Field Validation • Burn severity was measured in the field • Measurements compare well to change in roughness Sources of Error • Accuracy of individual datasets • Resolution of individual datasets • Accuracy of co-registration – Co-registration method – Degree of warping used – Unknown areas of change which bias the coregistration – Number of points used – Number of iterations Conclusions • LiDAR can be used effectively to detect and monitor change at the sub-meter level • LiDAR-based change detection can be used in a variety of environments • Statistical methods are useful for leveraging the large amounts of data in LiDAR studies • Care must be taken to preserve the highaccuracy of the raw LiDAR data Questions?